The report describes behaviour similar to that of a so-called "computer worm" - malicious software that spreads between computers via the internet on its own.
An AI worm could jump from device to device via networks of poorly protected devices in the home, such as set-top boxes or kitchen appliances. According to the report’s authors, it would likely take several days to copy itself to another device, similar to the behaviour of a swarm of fruit flies. Once the swarm is established, it can be difficult to eradicate because it does not rely on large data centres.
The researcher: The models are too slow
Sverker Janson, head of the Center for Applied AI at the research institute RISE, believes today’s local AI models are still too slow, limited and demanding to pose an immediate threat to ordinary users.
Individuals don’t need to worry about AI getting in and running on their computers or home products.
According to Sverker Janson, smart and large AI models would be needed to cause damage, and they cannot be accommodated in home products.
“Good to warn of different scenarios”
Although Sverker Janson urges calm, he believes it is still positive that Anthropic raises the risks of AI.
Anthropic is quite visionary in how they view AI risks and what might happen in the future. And it’s good that someone is, because there’s nothing today that contradicts how far AI’s capabilities will develop.
Anthropic, which developed the AI model Claude, has been accused by the White House and Silicon Valley business leaders of whipping up fear around AI to protect its own dominance against smaller competitors.
What do you think?